mirror of
https://github.com/fumiama/blake2b-simd.git
synced 2026-06-05 02:00:26 +08:00
cf4f8e0c34
``` $ go test -run=NONE -bench . PASS BenchmarkHash64-4 1000000 1036 ns/op 61.77 MB/s BenchmarkHash128-4 2000000 801 ns/op 159.67 MB/s BenchmarkHash1K-4 500000 2464 ns/op 415.53 MB/s BenchmarkHash8K-4 200000 11212 ns/op 730.60 MB/s BenchmarkHash32K-4 30000 40766 ns/op 803.80 MB/s BenchmarkHash128K-4 10000 163170 ns/op 803.28 MB/s ok github.com/minio/blake2b-simd 10.298s ```
BLAKE2b-SIMD
Pure Go implementation of BLAKE2b using SIMD optimizations.
Introduction
This package is based on the pure go BLAKE2b implementation of Dmitry Chestnykh and merges it with the (cgo dependent) SSE optimized BLAKE2 implementation (which in turn is based on official implementation. It does so by using Go's Assembler for amd64 architectures with a fallback for other architectures.
It gives roughly a 3x performance improvement over the non-optimized go version.
Benchmarks
| Dura | 1 GB |
|---|---|
| blake2b-SIMD | 1.59s |
| blake2b | 4.66s |
Example performance metrics were generated on Intel(R) Core(TM) i7-3520M CPU @ 2.90GHz - 2 physical cores, 4 logical cores running Ubuntu GNU/Linux with kernel version 4.4.0-22-generic.
benchmark old ns/op new ns/op delta
BenchmarkHash64-4 742 411 -44.61%
BenchmarkHash128-4 681 346 -49.19%
BenchmarkWrite1K-4 4239 1497 -64.69%
BenchmarkWrite8K-4 33633 11514 -65.77%
BenchmarkWrite32K-4 134091 45947 -65.73%
BenchmarkWrite128K-4 537976 183643 -65.86%
benchmark old MB/s new MB/s speedup
BenchmarkHash64-4 86.18 155.51 1.80x
BenchmarkHash128-4 187.96 369.10 1.96x
BenchmarkWrite1K-4 241.55 683.87 2.83x
BenchmarkWrite8K-4 3897.06 11383.41 2.92x
BenchmarkWrite32K-4 977.48 2852.63 2.92x
BenchmarkWrite128K-4 243.64 713.73 2.93x
We can see 2-3x improvement in performance over native Go under varying block sizes.